This paper examines how pecuniary penalties under China's One-Child Policy (OCP) affected female educational attainment by changing expected marriage market gains. OCP fines on second and subsequent births increase the relative marriage market value of older women, encouraging delayed entry to capture greater labor market returns. However, because delayed entry creates unbalanced trade-offs between labor market gains and marriage timing across education levels, its net effect on schooling investments is theoretically ambiguous. We formalize this mechanism using an overlapping generations model with transferable utility where forward-looking women choose schooling and entry timing. Methodologically, we overcome the unobservability of early-stage single rates and test a novel co-movement condition that identifies directional shifts in expected marriage market returns with the National Population Census of China and newly constructed provincial OCP fine data. Among those affected (i.e., urban Han women), an increase in average expected OCP fines equivalent to one year of household income between ages 6 and 20 increases the probability of high school completion by more than 10 percentage points, whereas showing no significant placebo effect for rural counterparts facing weaker fertility constraints. Concurrently, it widens the later-stage matching probability gap between high school graduates and non-graduates by approximately 5 percentage points, translating into an expanding advantage in expected marriage gains for graduates. Our findings demonstrate that fertility-reducing policies shape female human capital accumulation not only by altering reproductive expectations, but also by reallocating gains within the marriage market.
This paper characterizes a class of individual preferences in which heterogeneous social members care not only about their own consumption, but also about the minimum consumption in society. The key axiom triggering such concern is an indifference preference on the consumption distribution of others whenever a social member is “miserable”, defined as possessing the lowest disposable endowment after transfer. The characterized individual preferences, represented by a linear combination of a personalized social utility function increasing in the minimum consumption in society and an egoistic utility function increasing in one's own consumption, can support an endogenous social minimum without relying on any exogenously-given judgement. This paper then evaluates the dynamics of the endogenous social minimum by the ruling of a benevolent utilitarian social planner, and reveals certain path-dependent feature of the redistribution process. While a social planner can facilitate higher transfers from rich social members to poor social members and minimize the consumption inequality, the endogenous social minimum can also be supported through voluntary contributions from social members following the common approach in public goods literature. Nevertheless, total voluntary contribution will converge to zero if the society expands by self-replicating the original set of social members.
Boston College IRB Review Exempt Approval - Protocol Number: 25.1046
In this paper, we report results from a field study that examines the effects of nudging students to use generative AI study tools by providing free access to the premium Chegg AI platform. The platform supports parallel use of several leading foundation models, including OpenAI’s GPT models, Anthropic’s Sonnet models, and Google’s Gemini models, along with Chegg’s proprietary large language model (LLM), which is fine-tuned specifically on educational content and designed to select the "best" answer across models. Crucially, we find no evidence that nudging students to use GenAI by providing free access to the premium tool harmed learning outcomes: the treated section’s final exam performance was at least as good as that of the control section. However, the benefits of the nudge were highly heterogeneous. Students with stronger prior performance, those who initially held more positive attitudes towards AI, those who felt more confident with the material, and those who typically avoided asking instructors and teaching assistants for help realized the largest gains. By contrast, students who were skeptical of AI or heavily reliant on instructor support saw little to no improvement. These patterns point to complementarity between the AI tools and students’ prior skill and confidence in the material, alongside substitution between AI use and proactive in-class help-seeking. The nudge also altered study habits: treated students reported spending roughly two fewer total study hours per week on the course, yet achieved comparable final exam performance. In addition, heavier use of generative AI observed among treated students is associated with lower instructor and course evaluations and more favorable attitudes toward AI. Together, these findings suggest that greater use of generative AI substitutes for both conventional study effort and human engagement. Our findings highlight both the promise and the pitfalls of integrating generative AI into education. AI tools can boost productivity without sacrificing learning, but they disproportionately benefit higher-skilled, more "AI-ready" students and may substitute for valuable human interactions and student effort in ways that could hinder longer-term skill development.
Banner Photo by Fatih Gezer